Metal target detection method using passive millimeter-wave polarimetric imagery

Fangzhou Tang; Liangqi Gui; Liangqi Gui; Jinbang Liu; Ke Chen; Ke Chen; Liang Lang; Liang Lang; Yayun Cheng

doi:10.1364/OE.390385

1. Introduction

Passive millimeter-wave imaging (PMMW) has been developed as a useful method for various applications in the terrestrial remote sensing and security industry due to their excellent ability to penetrate smoke, clothing, storms [1–4]. PMMW imaging is based on the passive detection of naturally occurring millimeter-wave radiation. The numerical values of apparent temperature determined by the emissivity, the physical temperature, and the ambient temperature are displayed in each pixel. With favorable penetrability, in both low- and high-visibility conditions, PMMW imaging is an excellent tool for metal target detection in a complex background. The different electromagnetic properties between metal target and typical background can cause differences in the emission and reflection of millimeter-wave. These differences increase the contrast between each pixel in the PMMW image. Therefore, PMMW images have potential use in target detection.

Nevertheless, millimeter-wave radiation of the target surface is sensitive to the observation angle and factors which affect the roughness. The emissions of other natural targets that often occur in the scene also have effects. These factors sometimes have led to an inability to distinguish metal target and background directly in millimeter-wave images in complicated scenes. The polarization properties of PMMW can capture inherent information about targets, such as composition and surface features. Therefore, polarimetric measurement has been widely used to acquire additional information, such as the index of refraction, reflectivity, orientation, and material classification [5–8]. In infrared and microwave or millimeter-wave regions, brightness temperatures, Stokes parameters, and degrees of polarization are widely utilized [9–11]. The polarization phenomena and recognition technology developed based on Stokes parameters have been successfully applied to systems such as atmospheric remote sensing, aerial photography, and military target recognition [12–15]. Similar to these parameters, degree of linear polarization, degree of circular polarization, multispectral polarimetric methods, and passive degree of polarization (PDoP) are also claimed to be effective [7]. In addition, the linear polarization ratio (LPR) has been suggested to be utilized at an optimal incident angle [5]. Furthermore, polarization brightness temperatures are used for measuring large ice sheets and wind fields in areas of target detection on sea surfaces [14,15]. Yujiri demonstrated PMMW imaging of stationary and moving metal ship [16]. The contrast between the metal ship and the background is evident. Traditional non-polarimetric method mainly rely on the difference in the brightness temperature between target and background to detect the target [17]. However, due to the dramatic variability of ambient brightness temperatures, when the environment changes, the brightness temperature of the target metal will considerably varies such that the background cannot be differentiated. The metal brightness temperature in some directions may be higher than the background. Polarization is measured based on the law between the brightness and the temperature of the material and is not affected by the environment, which can help avoid the background problem. Therefore, one-dimensional apparent temperature information cannot accurately distinguish metal targets. In addition to general radiation measurment approaches, polarimetric electromagnetic theory and statistical analysis might be a more promising solution.

To identify a metal target by using PMMW images, the detection criteria need to be sensitive to the metal target type and must eliminate the confounding effects of the ambient radiation and incident angles. The feature discriminator should not contain parameters of ambient radiation, and the classification results at each incident angle should be assured to be adequate. DLPD theory is presented here, and characteristics of typical natural targets are illustrated by theoretical and experimental analysis. LPDR is selected as a feature discriminator that maintains a near-perfect sensitivity to a metal target and typical environmental background, and it reduces the confounding effects of ambient radiation and the angle of observation. Outstanding detection results on metal targets and against various backgrounds under the proposed classification criteria are verified. Outdoor experiments are conducted with a 94-GHz-radiation imaging system. The measurement test proves to have a stable discrimination capacity in detecting a metal target in a complex background. The results indicate that the proposed detection technique can be implemented to detect a metal target in complicated environmental conditions and that the resulting LPDR image is quite visible.

2. Dual linear polarization discriminator theory

2.1 Fundamental theory

Thermal radiation theory states that all media (gases, solids, liquids, and plasmas) emit radiation in the form of discrete energy as functions of their absolute temperature. The radiation that is received by an antenna from any direction is made up of several different components which can be divided into surface emission and the reflection of ambient radiation. The radiance emitted in millimeter-wave bands can be approximated as a linear function of physical temperature by the Rayleigh-Jeans approximation [10]. The reflection of ambient radiation can be derived from the Kirchhoff law [18]. The emissivity $e$ is equal to the absorbance at the local thermodynamic equilibrium. For a homogeneous, isothermal, "opaque" medium, the brightness temperature of a flat surface at a specific azimuth angle can be described by Eq. (1) [19],

(1)$${T_B}(\theta ;\varphi ;p) = e(\theta ;p){T_{obj}} + [1 - e(\theta ;p)]{T_{Binc}}(\theta ;\varphi ),$$

where $\theta$ is the observation angle; $\varphi$ is the azimuth angle; $p$ represents the direction of polarization. $e$ and $\Gamma$ are the millimeter-wave emissivity and reflectivity of the target surface ($e=1-\Gamma$). $T_{obj}$ is the target physical temperature, and $T_{Binc}$ is the ambient brightness temperature incident on the target at the angle of reflection. Target brightness temperature is independent of azimuth angle, and the incident environmental radiation is related to the azimuth angle. In previous polarization theory, the Stokes parameters are used to describe the intensity and polarization of the optical wave. Since the polarization parameter is used as the imaging factor, it records more information than just the single value of intensity. In PMMW polarimetric remote sensing, Stokes parameters are commonly used for substance parameter retrieval and target recognition. [20,21]. In radiation metrology, Stokes parameters can be written as polarization brightness, as Eq. (2),

(2)$$\overline {{T_B}} = \left( {\begin{array}{c} {{T_I}}\\ {{T_Q}}\\ {{T_U}}\\ {{T_V}} \end{array}} \right) = \left( {\begin{array}{c} {({T_{Bv}} + {T_{Bh}})/2}\\ {{T_{Bv}} - {T_{Bh}}}\\ {{T_{B{{45}^{{\circ}} }}} - {T_{B - {{45}^{{\circ}} }}}}\\ {{T_{Bcl}} - {T_{Bcr}}} \end{array}} \right),$$

where $\overline {{T_B}}$ is the radiation apparent brightness temperature. $T_{Bv}$, $T_{Bh}$ respectively are used to describe the horizontal polarization and vertical polarization brightness temperature. $T_U$, $T_V$ are respectively used to describe the linear polarization and circular polarization. $T_k$ is the brightness temperature, with the subscript $k$ being the different polarization modes: linear = $+45^{\circ }$ and $-45^{\circ }$, circular left = $cl$ and circular right = $cr$, respectively. The azimuth is referenced to the target surface, not the horizontal ground. Stokes parameters completely describe the polarization characteristics of microwave radiation. Assuming that the incident radiation from the environment is non-polarized, $T_I$, $T_Q$, dual polarization, which can be written as Eq. (3),

(3)$$\left( {\begin{array}{c} {{T_{I(\theta ;\varphi )}}}\\ {{T_{Q(\theta ;\varphi )}}} \end{array}} \right) = \left( {\begin{array}{c} {[{T_{obj}} - {T_{Binc}}(\theta ;\varphi )][{e_v}(\theta ) + {e_h}(\theta )] + 2{T_{Binc}}(\theta ;\varphi )}\\ {[{T_{obj}} - {T_{Binc}}(\theta ;\varphi )][{e_v}(\theta ) - {e_h}(\theta )]} \end{array}} \right),$$

where ${e_h}(\theta )$, ${e_v}(\theta )$ denote horizontally and vertically polarized emissivity. $T_{Binc}$ is a variable parameter in a real environment. Thus, $T_I$ and $T_Q$ are related not only to the surface emissivity but also to ($T_{obj} - T_{Binc}$). In previous theory, the variability of $T_{Binc}$ will affect the accuracy of the evaluation [7]. To remove the influence of ambient radiation, DLPD is defined by the ratio of orthogonal polarization reflectivities as follows:

(4)$$DLPD(\alpha ,\beta ,\lambda ,\gamma ,\theta ) = \frac{{\alpha {\Gamma _v}(\theta ) + \beta {\Gamma _h}(\theta )}}{{\lambda {\Gamma _v}(\theta ) + \gamma {\Gamma _h}(\theta )}},$$

where $\alpha$, $\beta$, $\gamma$, $\theta$ are independent variable parameters. DLPD reveals the internal mechanism of the polarization characteristic quantity and has a variable form related to the dual linear polarization matrix (DLPM). Four typical matrices can be given with two mathematical constraints: DLPD should exist and can remove the influence of ambient radiation.

(5)$$DLPM = \left[ {\begin{array}{cc} {\begin{array}{cc} \alpha & \beta \end{array}}\\ {\begin{array}{cc} \lambda & \gamma \end{array}} \end{array}} \right] \in A{*}\left\{ {\left[ {\begin{array}{cc} 1 & 0\\ 1 & 0 \end{array}} \right]\left[ {\begin{array}{cc} 1 & 0\\ 1 & { - 1} \end{array}} \right]\left[ {\begin{array}{cc} 1 & 0\\ 1 & 1 \end{array}} \right]\left[ {\begin{array}{cc} 1 & { - 1}\\ 1 & 1 \end{array}} \right]} \right\},$$

where $A$ is an arbitrary non-zero constant. Ignoring reciprocal and opposite equations, four DLPMs are given in Eq. (5). The four matrices refer to different discriminators, respectively, as expressed in Eqs. (6)–(9):

(6)$$LPDR(\theta ) = \frac{{{\Gamma _v}(\theta )}}{{{\Gamma _h}(\theta ) - {\Gamma _v}(\theta )}} = \frac{{1 - {e_v}(\theta )}}{{{e_v}(\theta ) - {e_h}(\theta )}},$$

(7)$$DoSP(\theta ) = \frac{{{\Gamma _v}(\theta )}}{{{\Gamma _h}(\theta ) + {\Gamma _v}(\theta )}} = \frac{{1 - {e_v}(\theta )}}{{2 - [{e_v}(\theta ) + {e_h}(\theta )]}},$$

(8)$$PD{\textrm{oP}}(\theta ) = \frac{{{\Gamma _v}(\theta ) - {\Gamma _h}(\theta )}}{{{\Gamma _h}(\theta ) + {\Gamma _v}(\theta )}} = \frac{{{e_v}(\theta ) - {e_h}(\theta )}}{{2 - [{e_v}(\theta ) + {e_h}(\theta )]}},$$

(9)$$LPR(\theta ) = \frac{{{\Gamma _v}(\theta )}}{{{\Gamma _h}(\theta )}} = \frac{{1 - {e_v}(\theta )}}{{1 - {e_h}(\theta )}}.$$

In previous work, the detailed analysis of LPR and PDoP has proved that these two methods are useful in material classification and clustering [5,7]. In this paper, the LPDR and the degree of sole polarization (DoSP) are analyzed and compared, using theory and experimental data. Previous equations only described the theoretical relationship between DLPD and the polarization reflectivities. In actual measurements, LPDR and DoSP need to be calculated based on the measured brightness temperature. Under the assumption that the incident radiation from the environment is non-polarized, LPDR and DoSP can be obtained according to Eq. (10),

(10)$$\left\{ {\begin{array}{c} {{e_h}(\theta ) = \frac{{{T_{Bh}}(\theta ;\varphi ) - {T_{Binc}}(\theta ;\varphi )}}{{{T_{obj}} - {T_{Binc}}(\theta ;\varphi )}}}\\ {{e_v}(\theta ) = \frac{{{T_{Bv}}(\theta ;\varphi ) - {T_{Binc}}(\theta ;\varphi )}}{{{T_{obj}} - {T_{Binc}}(\theta ;\varphi )}}} \end{array}} \right..\,$$

By substituting Eq. (10) into Eq. (6) and Eq. (7), measurement formulas for LPDR and DoSP can be calculated by Eq. (11) and Eq. (12).

(11)$$LPDR(\theta ) = \frac{{{T_{obj}} - {T_{Bv}}(\theta )}}{{{T_{Bv}}(\theta ) - {T_{Bh}}(\theta )}},$$

(12)$$DoSP(\theta ) = \frac{{{T_{obj}} - {T_{Bv}}(\theta )}}{{2{T_{obj}} - [{T_{Bv}}(\theta )\textrm{ + }{T_{Bh}}(\theta )]}}.$$

2.2 Comparison of different discriminators

To investigate the DLPD characteristics of metal targets and other matter, polarization emissivities of surfaces must be calculated. By Fresnel’s law, assuming the materials are nonmagnetic, the radiation brightness temperature of flat, non-transparent targets is available [18]. Dielectric permittivity and conductivity in the millimeter-wave region are obtained from published articles [22–27]. Figure 1 depicts the polarization emissivities of many natural and artificial materials. Horizontal and vertical polarization emittances show different behaviors with changes in observation angle.

Fig. 1. Theoretical emissivity of several natural and artificial materials. (a) Horizontal polarization emissivity (b)vertical polarization emissivity. H and V denote horizontal and vertical polarization respectively.

Download Full Size | PDF

Figure 1 shows that the horizontal and vertical polarization emissivities of metallic materials are close to 0. On the contrary, horizontal and vertical polarization emissivities of the dielectric material are different. According to Eq. (1), the metallic material has a comparatively smaller difference between the horizontal and vertical polarization brightness temperature. Thus, the orthogonal polarization difference guarantees that LPDR is more sensitive to the type of material. Since emissivity changes with observation angle, $1-ev$ includes the influence of the angle of emission/observation. Furthermore, division eliminates $T_{Binc}$ in the Stokes vectors. According to the definition of DLPD, the 94-GHz DLPD results for several natural and artificial targets are shown in Fig. 2. The calculated results indicate significant differences in logarithmic LPDR for these two types of materials, metallic and dielectric. The red arrows in Fig. 2 indicate the distance between the DLPD values of metal and dielectric. The LPDR of typical metals (e.g., aluminum, copper) remains stable and always higher than the LPDR of dielectrics, under almost all observation angles. This is due to the difference in the polarization emissivity between metals and dielectrics. Hence, LPDR may be a new feature discriminator which can be used for metal target detection.

Fig. 2. Theoretical DLPD values of several natural and artificial materials. (a) LPDR (b) DoSP (c) PDoP (d) LPR.

Download Full Size | PDF

To verify the above theoretical analysis, measurements are carried out to obtain the DLPD values of several different materials. The radiometer used in these measurements is 94 GHz radiometer with 2 GHz bandwidth and 10ms integration time. The 3 dB beam width of the radiometer antenna is $0.8^{\circ }$. The radiometric sensitivity of the radiometer antenna is approximately 0.8K. The range of incident angle is $45^{\circ }\sim 80^{\circ }$. The red arrows in Fig. 3 indicate the distance of DLPD values of metal and dielectric. As shown in Fig. 3, the LPDR difference between the dielectrics and the metal is clear under different incident angles. On the contrary, the DoSP, PDoP, and LPR difference between the dielectrics and the metal under incident angle of $65^{\circ }\sim 70^{\circ }$ are larger than that under other incident angles. Therefore, LPDR is more stable than DoSP, PDoP, and LPR under different incident angles. This is consistent with the theoretical analysis in Fig. 2. The LPDR value of materials in the measurements have some difference with our theoretical calculations. This is because surface roughness and antenna geometry can influence LPDR in practical measurements, while these factors have not been considered in the theoretical calculations. Nonetheless, the differences between the calculated and measured responses have little effect on LPDR’s ability to distinguish dielectrics and metal. Owing to such excellent performance, with a large numerical range and robust angular characteristics shown in both the theoretical analysis and the experimental data, we selected LPDR as a method for target detection.

Fig. 3. DLPD measurement results of several natural and artificial targets using 94 GHz radiometer. All surfaces of targets are horizontal. (a) LPDR (b) DoSP (c) PDoP (d) LPR.

Download Full Size | PDF

3. Metal target detection simulation results

3.1 LPDR imaging simulation results

In order to identify metal targets in PMMW images by the LPDR properties of different materials, LPDR imaging results from different materials are simulated. The ray-tracing method is used to simulate the horizontal and vertical polarization images of a given scene [28]. As shown in Fig. 4, nine material plates (0.3 m $\times$ 0.3 m) of nine types (ceramic, silicon, soil, skin, pure water, seawater, ethyl alcohol, aluminum, and copper) are put on a concrete ground. All targets are illuminated with unpolarized sky radiation, and the material surfaces are assumed to be smooth in the millimeter-wave region. The observation angle is alterable (for instance, $60^{\circ }$ with respect to ground), and the observation distance between the radiometer and the scene center is 10 m. In practical applications, material surfaces may not always be strict horizontal. But for flat material, this effect has little influence on the incident angle and could be ignored.

Fig. 4. Simulation scene setup. (a) Planform of simulation scene (b)Side view of simulation scene.

Download Full Size | PDF

The dielectric permittivity and conductivity are obtained from published articles [22–27]. The physical temperature ($T_{obj}$) of all materials is set to 300 K. The 3 dB beam width of the radiometer antenna is $0.8^{\circ }$. The moisture of the soil is $5\%$. The salinity of sea water is $13\%$. The random Gaussian fluctuation (mean is 0 and variance is 0.8.) is added to brightness temperature images. Figure 5 depicts the simulation brightness temperature images at 94 GHz and the corresponding LPDR results at an incident angle of $60^{\circ }$. In Fig. 5, the nine materials are the same materials as shown in Fig. 4. Figures 5(a) and 5(b) show horizontal and vertical polarization brightness temperature images, respectively. For each material, the horizontal polarization brightness temperature is lower than the vertical polarization brightness temperature, which is consistent with the results shown in Fig. 1. Furthemore, metals and dielectrics have different brightness temperature characteristics. Since the emissivity of a metal is approximately zero, its brightness temperature is primarily observed from the reflection of ambient radiation. In the simulation, the metal reflects the radiation of the sky that is lower than that of the dielectric media. Therefore, the brightness temperature of metal is lower than that of the dielectric media. The brightness temperature of metals (copper and aluminum) shows little difference between the horizontal polarization and the vertical polarization. The vertical polarization brightness temperature of the dielectric is higher than the horizontal polarization brightness temperature. This is in accord with our theoretical calculation.

Fig. 5. Simulation images at 94 GHz and corresponding LPDR results. (a) Brightness temperature image of horizontal polarization (b) Brightness temperature image of vertical polarization (c) Ideal LPDR image (d) LPDR image considering antenna pattern and random noise fluctuation.

Download Full Size | PDF

However, it can be found that the brightness temperatures of ceramic and silica will be lower than that of other dielectrics, which are even closer to metal. Therefore, it is difficult to distinguish metal and dielectric through differences in their brightness temperature distributions. According to Eq. (8), LPDR values can be calculated with $T_{Bh}$, $T_{Bv}$, and $T_{obj}$. Figures 5(c) and 5(d) show ideal LPDR image and LPDR image considering antenna pattern and random noise fluctuation respectively. The clarity of the target in Fig. 5(d) is lower than that in Fig. 5(c). In addition, Figure 5 shows that the brightness temperature and LPDR characteristics of different materials are significantly different. Besides, the brightness temperature and LPDR values of different area in same material also have slight different. To quantitively observe the brightness temperature and LPDR characteristics of nine materials in Fig. 5, their average values are calculated and listed in Table 1. $T_{Bh}$ and $T_{Bv}$ are the brightness temperature of horizontal and vertical polarization of materials, respectively. LPDR (ideal) are the values of the LPDR image with ideal resolution. LPDR (antenna) are the values of the LPDR image considering the antenna pattern and random noise fluctuation. From Table 1 ,it is clear that the horizontal and vertical polarization brightness temperatures of the metallic materials (aluminium and copper) have slight differences. However, horizontal and vertical polarization brightness temperatures of dielectric materials (cermaic, pure water, silica, sea water, soil, skin, and ethyl alcohol) have larger difference than those of metallic materials. The LPDR values of the metallic materials are much larger than those of other materials. This is consistent with our theoretical calculations in Fig. 2(a).

Table 1. Average Values of Four Parameters

View Table | View all tables in this article

3.2 LPDR target detection criterion description

A specific LPDR criterion is analyzed here for the detection of a metal target. Noting that values of LPDR vary over a large range, logarithmic transformation can be utilized to describe the pixel distribution behavior. The classification criterion should be automatic so that metal could be distinguished from dielectrics in all LPDR images. To study the distribution characteristics of the LPDR images, histogram analysis is introduced. It is accepted that a histogram is especially useful when studying relative measurements. The PMMW brightness temperature image needs to be calibrated which makes the histogram analysis unsuitable. However, the DLPD theory constructs the polarization discriminator which uses the difference in polarization parameters to detect a target. The PMMW brightness temperature images are utilized to acquire the DLPD image. According to the DLPD theory, Eq. (6) and Eq. (10) are combined to obtain Eq. (11). Eq. (6) is the polarization discriminator and Eq. (10) contains the brightness temperature and polarization parameters. Thus, Eq. (11) expresses the polarization discriminator in terms of brightness temperature. Therefore, the LPDR polarization discriminator is also a relative measurement. The histogram analysis is justified for LPDR target detection criterion.

Figure 6 depicts the logarithmic histogram of LPDR images considering antenna pattern and random noise fluctuation under four incident angles. At various angles, the LPDR results of metals are similar and those of dielectrics are similar. However, the results for metals are considerably different from those of the dielectrics. This difference corresponds to a maximum distance: $D_{max}$. By determining $D_{max}$, the threshold can be determined.

Fig. 6. The corresponding histograms of logarithmic LPDR image considering antenna pattern and random noise fluctuation under four incident angles. (a) Under $25^{\circ }$ incident angle (b) Under $40^{\circ }$ incident angle (c) Under $60^{\circ }$ incident angle (d) Under $75^{\circ }$ incident angle.

Download Full Size | PDF

Based on LPDR characteristic of metals and dielectrics, an automatic detection criterion is proposed. The detection criterion is that material with LPDR $\geq$ $\delta$ represents metal, while the value of LPDR $<$ $\delta$ implies the material is dielectric ($\delta$ is the LPDR automatic threshold). The logarithmic pixel values of LPDR image, $P_i$, are sorted from minimum to maximum to get the array ${P_i}(j)$. $D(k)$ is the array of the difference values between adjacent elements of ${P_i}(j)$. In addition, due to the influence of the system noise, there will very few and irregularly distributed points at the bottom of the histogram of LPDR image. Considering a priori information derived from performance of imaging system: the proportion of target pixels in image can not be less than $0.05\%$. Therefore, pixel values whose proportion is less than $0.05\%$ will be treated as "system noise" and skipped. By finding the maximum difference $D(x)$, the value of $\delta$ is given by Eq. (13). The results of this detection method under four typical incident angles are shown in Fig. 7.

(13)$$\left\{ {\begin{array}{c} {D(x) = Max\{ {P_i}(j) - {P_i}(j - 1)\} }\\ {\delta = {\textrm{P}_i}(x) + \frac{{D(x)}}{2}} \end{array}.} \right.$$

Fig. 7. The detection results by using LPDR character in four incident angles. Black region donates metal target while white region donates background. (a) Under $25^{\circ }$ incident angle (b) Under $40^{\circ }$ incident angle (c) Under $60^{\circ }$ incident angle (d) Under $75^{\circ }$ incident angle.

Download Full Size | PDF

Figure 7 shows that the automatic threshold is effective in distinguishing the metal target from the background demonstrating the reliability of the automatic threshold under different incident angles. It takes into account the correlation between the pixels by calculating the difference in the adjacent values using Eq. (13). Therefore, the distribution characteristics of the pixels are considered by the automatic threshold technique. From Fig. 6, it is evident that the $D_{max}$ and $\delta$ in Figs. 6(a) and 6(d) are suitable for each image. Therefore, the automatic threshold is effective in LPDR target detection.

4. Metal target detection experiment

4.1 Measurement surrounding

In order to verify effective practical application of the proposed target detection technique, outdoor experiment in real scenes has been conducted. The detection of a metal ship target on the sea surface is widely considered important in the target detection field. Metal ship target surfaces have many different orientations such as vertical and horizontal surfaces. The background is complex in a natural environment. Hence, the metal ship can be selected as the metal target. The experimental setup is shown in Fig. 8. Imaging radiometer used is a 94-GHz scanning imaging radiometer shown in Fig. 8(a) with 2 GHz bandwidth and 10ms integration time. The 3 dB bandwidth of the radiometer antenna is $0.8^{\circ }$. The radiometric sensitivity of the radiometer antenna is approximately 0.8 K. In Fig. 8(b) The incident angle of image center is around $65^{\circ }$, and distance between radiometer and metal target is about 35 m. The metal ship target which has vertical and horizontal surface relative to the ground is shown in Fig. 9(a). Figure 9(b) is the optical image of imaging scene. In Fig. 9(b), the test scene contains four possible dielectrics in background: water, vegetation, concrete, and soil. There is one metal ship target, which has a larger observed angle. The sky is clear. By rotating the receiver to change the polarization receiving mode of the antenna, vertical and horizontal polarization images are obtained in different time respectively.

Fig. 8. Experimental setup and scene. (a) Experimental setup (b) Experimental scene.

Download Full Size | PDF

Fig. 9. Optical image of experimental scene. (a) Metal ship target (b) Imaging scene.

Download Full Size | PDF

Imaging results are shown in Fig. 10. Figures 10(a) and 10(b) show brightness temperatures of vertical polarization and horizontal polarization, respectively. It can be seen from Figs. 10(a) and 10(b) that the background vertical polarization brightness temperature is higher than the background horizontal polarization brightness temperature. The target vertical polarization brightness temperature is close to the target horizontal polarization brightness temperature. This is consistent with our theoretical calculations. Additionally, data markers have been added to Figs. 10(a) and 10(b) to indicate the brightness temperature of the target and background. Besides, it can be observed that the brightness temperature of the metal target is quite different from that of the background. In Fig. 10(b), there is a small difference between the nearby metal target and the water surface, so it is difficult to directly recognize the observed targets through the brightness temperature.

Fig. 10. Imaging results. (a) 94GHz horizontal polarization brightness temperature image (b) 94GHz vertical polarization brightness temperature image (c) LPDR Image.

Download Full Size | PDF

By fusing vertical polarization and horizontal brightness temperatures images using Eq. (11), LPDR image is obtained. Figure 10(c) is the LPDR image result. The target is clearer in LPDR image by inspection. In addition, in Figs. 10(a) and 10(b), the target seems to be clear and the metal target could perhaps be found with a detection criterion by single brightness temperature. This is reasonable because the experiment was conducted in an ideal environment: (1) In experimental scene shown in Fig. 8, the observation angle is around $65^{\circ }$. According to Fig. 1, $65^{\circ }$ is an observation angle under which target and background have an high contrast in brightness temperature. Therefore, the target is clear; (2) The sky is clear in Fig. 9(b). For metal target, the ambient radiation is manily from sky radiation. The sky radiation is lower than the materials in backgound. Therefore, the target is apparent in the brightness temperature image; (3) According to Fig. 10, there are many pixels in the target region. In the histogram, the counts of target will be evident which is helpful for detecting the target.

Detection results are shown in Fig. 11. The corresponding histogram of LPDR image is shown in Fig. 11(a). According to the automatic threshold proposed, the $D_{max}$ and $\delta$ are calcluated. The simulation results and experimental results appear different in some degree. There are mainly two reasons behind the differences in these results: (1) The simulated environment is simpler, and the experimental image environment is more complicated. The noise impact of hardware equipment is also very evident. This would be reflected on the histogram by a small amount of noise at the bottom of the image; however, the pixel values are widely distributed, as shown by the red ellipse in Fig. 11(a). (2) The $D_{max}$ appears small because the proportion of the target pixels is very small, resulting in a small number of pixels at the target on the right in the experimental image. The boundary of $D_{max}$ on the target is not obvious, as shown by the yellow ellipse in Fig. 11(a). In principle, these difference are caused by the complexity of the environment and system performance, and exist objectively. Figure 11(b) depicts the detection result of the metal ship target using the criterion given. The black region denotes the metal ship target. The white region denotes the background. Both simulations and experiments have demonstrated the effectiveness of our proposed target detection technique.

Fig. 11. Detection results. (a) The corresponding histogram of LPDR image (b) Target detection result.

Download Full Size | PDF

4.2 Image comparison and limitation analysis

In target detection, image quality can have a huge influence on detection accuracy. Further discussion about image quality is given here. Four kinds of images (LPDR, PDoP, DoSP, and LPR) are shown in Fig. 12. Figure 12 indicates that LPDR provided a good view of the imaging scene. In metal target detection, the contrast between the background and target is an essential factor. A higher contrast is more helpful in target detection. To evaluate this, there is no undistorted reference radiative image, so the image evaluation is done through an unreferenced image method. Four kinds of images are shown in Fig. 12. Figure 13 is the corresponding histograms respectively. The range of pixel value of target and background is marked. As shown in Figs. 13(a) and 13(b), the values of target are distributed away from the values of background. On the contrary, the values of target are distributed closer with the values of background in Figs. 13(c) and 13(d). A larger distance between target and background pixel is more helpful for detecting. It can be inferred that the contrast between target and background in LPDR and LPR image is higher than the contrast in DoSP and PDoP image.

Fig. 12. Four kinds of image. (a) LPDR image (b) DoSP image (c) PDoP image (d) LPR image.

Download Full Size | PDF

Fig. 13. The corresponding histogram of four kinds of image. (a) The corresponding histogram of LPDR image (b) The corresponding histogram of LPR image (c) The corresponding histogram of DoSP image (d) The corresponding histogram of PDoP image.

Download Full Size | PDF

To quantitatively compare the images’ contrast as obtained by different polarizations, the range of images pixel value is normalized so as to range from 0 to 255. According to the detection result of the target region in Fig. 11(b), the target and the background region are divided in each image. In Fig. 12, the target region is marked with red frame. Other area in image is regarded background region. Then the average of the pixel of the target region and the pixel values of the background region is calculated. The average values of pixel values of target and background are calculated and listed in Table 2. The contrast is the difference of average values of target pixels and average values of background pixels.

Table 2. The Comparison of Contrast in Different Images

View Table | View all tables in this article

In Table 2, contrast between target and background in different images is shown. The contrasts in LPR and LPDR image are higher than contrast in PDoP and DoSP image. Moreover, it seems that the contrast in LPR image is close to the contrast in LPDR image. This is because the complexity of background is ignored. All areas except target region are regarded as background region. In fact, there are vegetation, soil, water and concrete in background as shown in Fig. 9(b). LPDR and LPR also have different responses to each kind of background. Thus, it seems that the contrast in LPR image is close to the contrast in LPDR image. Although, Table 2 shows that LPDR image has better contrast.

Further experiments could make some improvements in three aspects. First, with a more flexible equipment, the experiment under multiple incident angles can be conducted and compared with current work. LPDR imaging data under multiple angle is valuable for evaluating the performance of LPDR method. Second, in real application, the background and targets are more complicated in type and structure. Different experiment conditions are needed to provide more data. Likewise, there are some limitations to the LPDR method in the MMW region. (1) Polarization measurements need to be accurate and adequate. Every pixel at one solid angle should be simultaneously obtained. (2) Ambient radiation should be unpolarized. The complex structure of real targets will cause rotation of polarization inside the target. But for a flat and large target, detection is effective, as our ship detection experiments showed. (3) The low resolution of the radiation image may decrease the number of the target’s pixels which will make it more difficult to detect metal target.

5. Conclusion

A metal target detection technique, DLPD, based on passive millimeter-wave polarimetric imagery, has been proposed and demonstrated. The feature discriminator, LPDR, which is sensitive to target type, is introduced to remove the interfering effects of reflected ambient unpolarized radiation and varied observation angles. Theoretical and experimental analysis has provided significant information about the LPDR characteristics of common materials. Simulations of some typical outdoor scenes have been conducted to validate the proposed detection technique. Moreover, because there are clear differences in LPDR in complex realistic situations, detection criteria is given.

Outdoor imaging experiment is conducted to validate the proposed target detection technique. The results show that the method proposed can identify a target in the imaging region and that better image quality can lead to possible further applications. More studies in threshold selection are necessary to improve the detection performance of LPDR. Some pixels are judged erroneously due to the complicated structure of the target and the complicated environment radiation, as we discussed above. The significant focuses of applying the proposed technique to complex scenes will be as follows: (1) The choice of the detection scheme parameters in different applications (e.g., edge detection and image segmentation); (2) The geometry and materials of construction of the target (e.g., three-dimensional reconstruction and attitude change of target); (3) The classification of different dielectric materials (e.g., liquid and solid).

Moreover, there are three significant concerns about the applicability of the LPDR technique. First, an appropriate range of observation angles is required. Observation angles will be more complicated in a practical situation. When the incident angle is close to $0^{\circ }$, the polarization difference of the background is small as that of the target. Second, detection at far distances, closer to the actual maritime application. The signal attenuation and environmental impact will make long-distance detection difficult. Third, the structure of the target will affect the accuracy of judgment. For instance, the polarization difference of a pit in the background is small and this may be misjudged as the target. Future studies should aim to address these concerns.

Funding

National Natural Science Foundation of China (61771213); Shanghai Aerospace Science and Technology Innovation Fund (SAST2017-113).

Acknowledgments

The authors would like to thank the editor and reviewers for their precious advice.

Disclosures

The authors declare no conflicts of interest.

References

1. Y. Su, S. Zhang, X. Shao, and J. Yang, “The dense filamentary giant molecular cloud G23.0–0.4: birthplace of ongoing massive star formation,” Astrophys. J. 811(2), 134 (2015). [CrossRef]

2. G. Hu, K. Chen, Q. Huang, W. Guo, Q. Li, L. Gui, and Y. Cheng, “Brightness temperature calculation of lunar crater: Interpretation of topographic effect on microwave data from Chang’E,” IEEE Trans. Geosci. Remote Sensing 52(8), 4499–4510 (2014). [CrossRef]

3. S. Yeom, D.-S. Lee, J.-Y. Son, M.-K. Jung, Y. Jang, S.-W. Jung, and S.-J. Lee, “Real-time outdoor concealed-object detection with passive millimeter wave imaging,” Opt. Express 19(3), 9371 (2012). [CrossRef]

4. R. Appleby and H. B. Wallace, “Standoff detection of weapons and contraband in the 100 GHz to 1 THz region,” IEEE Trans. Antennas Propag. 55(11), 2944–2956 (2007). [CrossRef]

5. F. Hu, Y. Cheng, L. Gui, L. Wu, X. Zhang, X. Peng, and J. Su, “Polarization-based material classification technique using passive millimeter-wave polarimetric imagery,” Appl. Opt. 55(31), 8690–8697 (2016). [CrossRef]

6. Y. Cheng, F. Hu, L. Gui, L. Wu, and L. Lang, “Polarization-based method for object surface orientation information in passive millimeter-wave imaging,” IEEE Photonics J. 8(1), 1–12 (2016). [CrossRef]

7. J. Su, Y. Tian, F. Hu, Y. Cheng, and Y. Hu, “Material clustering using passive millimeter-wave polarimetric imagery,” IEEE Photonics J. 11(1), 1–12 (2016). [CrossRef]

8. Y. Cheng, F. Hu, H. I. Wu, P. Fu, and Y. Hu, “Multi-polarization passive millimeter-wave imager and outdoor scene imaging analysis for remote sensing applications,” Opt. Express 26(16), 20145–20159 (2018). [CrossRef]

9. A. J. Yuffa, K. P. Gurton, and G. Videen, “Three-dimensional facial recognition using passive long-wavelength infrared polarimetric imaging,” Opt. Express 16(25), 8514 (2014). [CrossRef]

10. J. P. Wilson, C. A. Schuetz, T. E. Dillon, D. L. Eng, S. Kozacik, and D. W. Prather, “Display of polarization information for passive millimeter-wave imagery,” Opt. Eng. 51(9), 091607 (2012). [CrossRef]

11. J. R. Shell, “Polarimetric remote sensing in the visible to near infrared,” Ph.D.thesis, Rochester Institute of Technology (2005).

12. H. Chen and L. B. Wolff, “Polarization phase-based method for material classification in computer vision,” Int. J. Comput. Vision 28(1), 128–135 (1998). [CrossRef]

13. M. W. Hyde, S. C. Cain, J. D. Schmidt, and M. J. Havrilla, “Material classification of an unknown object using turbulence-degraded polarimetric imagery,” IEEE Trans. Geosci. Remote Sensing 49(1), 264–276 (2011). [CrossRef]

14. S. H. Yueh, W. J. Wilson, F. K. Li, S. V. Nghiem, and W. B. Ricketts, “Polarimetric measurements of sea surface brightness temperatures using an aircraft K-band radiometer,” IEEE Trans. Geosci. Remote Sensing 33(1), 85–92 (1995). [CrossRef]

15. O. Stahli, C. Matzler, A. Murk, and N. Kampfer, “A Surface-Based Imaging Method for Water Vapor and Liquid Clouds Using a Scanning Radiometer at 91 GHz,” IEEE Trans. Geosci. Remote Sensing 49(9), 3273–3280 (2011). [CrossRef]

16. L. Yujiri, M. Shoucri, and P. Moffa, “Passive millimeter-wave imaging,” IEEE Microw. Mag. 4(3), 39–50 (2003). [CrossRef]

17. Y. Meng, A. Qing, C. Lin, J. Zang, Y. Zhao, and C. Zhang, “Passive millimeter wave imaging system based on helical scanning,” Sci. Rep. 8(1), 7852 (2018). [CrossRef]

18. F. T. Ulaby, R. K. Moore, and A. K. Fung, Microwave Remote Sensing: Microwave Remote Sensing Fundamentals and Radiometry (Addison-Wesley Publishing Company, 1981), Vol. 1.

19. Y. Cheng, F. Hu, Y. Hu, Z. Zhao, and Y. Wang, “C-curve feature of complex permittivity estimation based on multi-polarization measurement in passive millimeter-wave sensing,” Opt. Lett. 44(15), 3765–3768 (2019). [CrossRef]

20. A. Duric, A. Magun, A. Murk, C. Matzler, and N. Kampfer, “The fully polarimetric imaging radiometer spira at 91 GHz,” IEEE Trans. Geosci. Remote Sensing 46(8), 2323–2336 (2008). [CrossRef]

21. P. S. Narvekar, G. Heygster, T. J. Jackson, R. Bindlish, G. Macelloni, and J. Notholt, “Passive polarimetric microwave signatures observed over Antarctica,” IEEE Trans. Geosci. Remote Sensing. 48(3), 1059–1075 (2010). [CrossRef]

22. H. Hwang, J. Yim, J. Cho, C. Cheon, and Y. Kwon, “110 GHz broadband measurement of permittivity on human epidermis using 1 mm coaxial probe,” in Proceedings of IEEE MTT-S International Microwave Symposium Digest (IEEE, 2003), pp. 399–402.

23. J. W. Lamb, “Miscellaneous data on materials for millimeter and submillimeter optics,” Int. J. Infrared Millimeter Waves 17(12), 1997–2034 (1996). [CrossRef]

24. H. J. Liebe, G. A. Hufford, and T. Manabe, “A model for the complex permittivity of water at frequencies below 1 THz,” Int. J. Infrared Millimeter Waves 12(7), 659–675 (1991). [CrossRef]

25. K. A. Korolev and M. N. Afsar, “Complex dielectric permittivity measurements of materials in millimeter waves,” in Proceedings of Joint 30th International Conference on Infrared and Millimeter Waves and 13th International Conference on Terahertz Electronics (IEEE, 2005), pp. 594–595.

26. K. Sato, H. Kozima, H. Masuzawa, T. Manabe, T. Ihara, Y. Kasashima, and K. Yamaki, “Measurements of reflection characteristics and refractive indices of interior construction materials in millimeter-wave bands,” in Proceedings of IEEE Vehicular Technology Conference (IEEE, 1995), pp. 634–642.

27. J. C. Calvet, J. P. Wigneron, A. Chanzy, S. Raju, and L. Laguerre, “Microwave dielectric properties of a silt-loam at high frequencies,” IEEE Trans. Geosci. Remote Sensing 33(3), 634–642 (1995). [CrossRef]

28. N. A. Salmon, “Polarimetric scene simulation in millimeter-wave radiometric imaging,” Proc. SPIE 5410, 260–269 (2004). [CrossRef]

Average Values	$T_{B h}$ (K)	$T_{B v}$ (K)	LPDR (ideal)	LPDR (antenna)
Cermaic	$108.9$	$111.8$	$1.785$	$1.784$
Aluminium	$89.55$	$89.65$	$3.405$	$3.402$
Pure water	$202.1$	$223.9$	$0.5405$	$0.5421$
Silica	$111.9$	$115.6$	$1.699$	$1.697$
Copper	$89.62$	$89.53$	$3.611$	$3.604$
Sea water	$202.3$	$224.2$	$0.5361$	$0.5408$
Soil	$263.7$	$278.5$	$0.1592$	$0.1469$
Skin	$239.3$	$259.1$	$0.3177$	$0.3127$
Ethyl alcohol	$272.5$	$285.3$	$0.06864$	$0.06207$

Values	LPDR	DoSP	PDoP	LPR
Target	$240$	$12$	$250$	$225$
Background	$77$	$37$	$216$	$63$
Contrast	$163$	$25$	$34$	$162$

Average Values	$T_{B h}$ (K)	$T_{B v}$ (K)	LPDR (ideal)	LPDR (antenna)
Cermaic	$108.9$	$111.8$	$1.785$	$1.784$
Aluminium	$89.55$	$89.65$	$3.405$	$3.402$
Pure water	$202.1$	$223.9$	$0.5405$	$0.5421$
Silica	$111.9$	$115.6$	$1.699$	$1.697$
Copper	$89.62$	$89.53$	$3.611$	$3.604$
Sea water	$202.3$	$224.2$	$0.5361$	$0.5408$
Soil	$263.7$	$278.5$	$0.1592$	$0.1469$
Skin	$239.3$	$259.1$	$0.3177$	$0.3127$
Ethyl alcohol	$272.5$	$285.3$	$0.06864$	$0.06207$

Values	LPDR	DoSP	PDoP	LPR
Target	$240$	$12$	$250$	$225$
Background	$77$	$37$	$216$	$63$
Contrast	$163$	$25$	$34$	$162$

Metal target detection method using passive millimeter-wave polarimetric imagery

Abstract

1. Introduction

2. Dual linear polarization discriminator theory

2.1 Fundamental theory

2.2 Comparison of different discriminators

3. Metal target detection simulation results

3.1 LPDR imaging simulation results

3.2 LPDR target detection criterion description

4. Metal target detection experiment

4.1 Measurement surrounding

4.2 Image comparison and limitation analysis

5. Conclusion

Funding

Acknowledgments

Disclosures

References

Cited By

Figures (13)

Tables (2)

Equations (13)

Optics Express